Real-time detection method for tiny object in high-speed image

The invention discloses a real-time detection method for a tiny object in a high-speed image, and the method comprises the steps: selecting to-be-recognized particle graphs with different attitudes from a detected image, obtaining a dictionary D related to the features of the to-be-recognized particle graphs through employing a dictionary learning method, and setting a convolution threshold T, andthe method for detecting the image based on the dictionary D comprises the following steps: S1, reading a first row of image data from the detected image, carrying out convolution on the first row ofimage data and a dictionary cache d, and comparing the maximum value of a convolution result with the convolution threshold T; S2, sequentially reading each row of residual image data in the detectedimage, sequentially carrying out convolution on each row of image data and the dictionary cache d, and comparing the maximum value of a convolution result with a convolution threshold T; and S3, if the maximum value of the con